Hate Speech Detection in Social Media (Twitter) Using Neural Network

Abstract

Hate speech recently became a real threat in social media, and almost all social media users are intended to in different ways. Hate speech is not limited to a group or society. It affects many people and can be classified as abusive, offensive, sexism, racism, political affiliation, religious hate, nationality, skin color, disability, gender-based, ethnicity, sexual orientation, immigrants, and others. Many researchers and authorities attempt to discover new procedures to sense hate speech in social media, especially on Facebook and Twitter, and many methods, models, and algorithms are used for this purpose. One of the most valuable models for detecting hate speech is Convolutional Neural Network (CNN). This review aims to assort academic studies on hate speech detection in Twitter using CNN-based models summarize the results of each model to expand the understanding of the recent circumstances of hate speech detection in Twitter. For this purpose, we implemented a broad, automated search using Boolean and Snowballing searching methods to find academic works in this area. Studies and papers have been distinguished, and the following information was obtained and aggregated from each article: authors, publication’s year, the journal name or the conference name, proposed model/method, the aim of the study, the outcome, and the quality of each study. According to the findings, the CNN and CNN-based models are standard models for hate speech detection. Besides, the findings show that other new models have a great compact on hate speech detection, and there is good progress in this field. However, the problems that still exist with hate speech detection models mainly are; most of the models cannot detect hate speech automatically. The methods are not suitable with all the languages, and they are working only with one language; most are best suited with the English language, and when they are used with datasets with other languages. Besides, the models are suffering from confusion in speech classification. Finally, most models are not considering a user-to-user speech in social media.